Stochastic Integer Programming: Parallel distributed-memory Algorithms

Wednesday, February 24, 2016 - 11:50am - 12:15pm
Lind 305
Deepak Rajan (Lawrence Livermore National Laboratory)
Many complex applications, including applications from electric power systems, can be modeled as stochastic optimization problems with integer decision variables. In stochastic optimization, the source of uncertainty in the model is usually approximated with a pre-defined number of possible realizations, called scenarios. Standard methods using extended formulations, known as deterministic equivalent Mixed-Integer Programs (MIPs), are often intractable when the number of scenarios is large.

In this talk, I will present two algorithmic approaches being developed at Lawrence Livermore National Laboratory that rely on decomposition at different levels of the algorithms. Both these algorithms are designed with parallelism in mind, and we present preliminary results using our algorithms on test instances from stochastic programming libraries.